Abstract
In a general normed vector space, we study the perturbed minimal time function determined by a bounded closed convex set \(U\) and a proper lower semicontinuous function \(f(\cdot )\). In particular, we show that the Fréchet subdifferential and proximal subdifferential of a perturbed minimal time function are representable by virtue of corresponding subdifferential of \(f(\cdot )\) and level sets of the support function of \(U\). Some known results is a special case of these results.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Let \(X\) be a normed vector space, \(U\) be a bounded closed convex subset of \(X\), and \(f{:}\,X\rightarrow \bar{R}\) be a proper lower semicontinuous function. We define the perturbed minimal time function \(T^f{:}\,X\rightarrow R\) by
where \(T(x,s):=\inf \{t\ge 0{:}\,s-x\in tU\}.\) It is easy to see that, if \(U\equiv B\), then \(T(x-s)=\Vert x-s\Vert \), where \(B\) is the unit ball in \(X\).
For \(x\in X\), the perturbed minimal time problem is to find an element \(z_0\in X\) such that
In particular, if \(f=I_S\), where \(I_S\) denote the indicator function \(I_S\) of a closed set \(S\) (the definition will be given below), then the perturbed minimal time function \(T^f\) reduces to the minimal time function \(T_S\) in [14], which is defined by the following differential inclusion
In other words,
If \(f=J+I_S\) and \(U\equiv B\), then the perturbed minimal time function \(T^f\) and the perturbed minimal time problem reduce to the perturbed distance function \(d_S^J\) and the perturbed optimization problem min\(_J(x,S)\) defined in [23], respectively, that is,
and
Baranger [1] proved that if \(S\) is a nonempty closed subset of a uniformly convex Banach space \(X\), then the set of all \(x\in X\) for the perturbed optimization problem min\(_J (x, S)\) has a solution is a dense \(G_\delta \)-subset of \(X\), which extends a result in [22] on the best approximation problem. For other results on perturbed optimization problems, see for example [3, 8, 9, 15, 16, 18–21]. In particular, Cobzas [9] extended Baranger’s result to the setting of reflexive Kadec Banach space; while Ni [18] relaxed the reflexivity assumption made in Cobzas’ result. The existence results have been applied to optimal control problems governed by partial differential equations, see for example, [1–3, 8, 12].
Assuming that the origin is an interior point of \(U\), Colombo and Wolenski [10, 11] studied the proximal and Fréchet subdifferentials of the function \(T_{S}(x)\) in a Hilbert space. He and Ng [13] studied the Fréchet and proximal subdifferentials of \(T_{S}\) in a Banach space. When the origin is an interior point of \(U\), the function \(T_{S}\) is globally Lipschitz, so the Clarke subdifferential of \(T_{S}\) is also discussed in [13]. Jiang and He [14] show the Frechét and proximal subdifferentials of the minimal time function \(T_{S}\) without requiring the origin be an interior point of \(U\) in normed space. In particular, if \(U\) is the (closed) unit ball in \(X\), then \(T_{S}(x)\) reduces to the usual distance \(d_S(x)\), which is defined by
The subdifferentials of \(d_S\) were studied in [4–7], and the subdifferentials of perturbed distance functions \(d_S^J\) were studied in [17, 23].
In order to reduce the symmetry of the norm, we replace the distance function in [23] by \(T(\cdot ,\cdot )\), which does not have the symmetry, and explore the Fréchet subdifferentials and the Proximal subdifferentials of its perturbed functions \(T^f(\cdot )\), the perturbed functions \(T^f(\cdot )\) are encountered in constraint optimization, via applying various perturbation, penalization, and approximation techniques. Our main results extend the corresponding ones in [14] from the minimal time function to perturbed minimal time function, and extend the corresponding ones in [23] from the general perturbed distance functions to general perturbed minimal time functions.
2 Preliminaries
Let \(X\) be a normed vector space with norm denoted by \(\Vert \cdot \Vert \). Let \(X^*\) denote the topological dual of \(X\). We use \(B(x;r)\) to denote the open ball centered at \(x\) with radius \(r>0\) and \(\langle \cdot ,\cdot \rangle \) to denote the pairing between \(X^*\) and \(X\). Let \(g{:}\,X\rightarrow \mathbb R \) be a lower semicontinuous function and \(x\in X\). \(g\) is said to be center Lipschitz on \(S\) at \(x\) with Lipschitz constant \(L\), if
Let us recall the following well-known classes of subdifferentials for \(g\) at \(x\).
-
The proximal subdifferential of \(g\) at \(x\) is the set
$$\begin{aligned} \partial ^Pg(x):=\left\{ \xi \in X^*{:}\,\liminf _{\Vert v\Vert \rightarrow 0}\frac{g(x+v)-g(x)-\langle \xi ,v\rangle }{\Vert v\Vert ^2}>-\infty \right\} . \end{aligned}$$In other words, \(\xi \in \partial ^Pg(x)\) if and only if there exist \(\sigma >0\) and \(\delta >0\) such that
$$\begin{aligned} g(x+v)-g(x)\ge \langle \xi ,v\rangle -\sigma \Vert v\Vert ^2,\quad \text {for all }v\in B(0,\delta ). \end{aligned}$$ -
The Frechét subdifferential of \(g\) at \(x\) is the set
$$\begin{aligned} \partial ^{F}g(x):=\left\{ \xi \in X^*{:}\,\liminf _{\Vert v\Vert \rightarrow 0}\frac{g(x+v)-g(x)-\langle \xi ,v\rangle }{\Vert v\Vert }\ge 0\right\} . \end{aligned}$$That is, \(\xi \in \partial ^Fg(x)\) if and only if for any \(\sigma >0\), there exists \(\delta >0\) such that
$$\begin{aligned} g(x+v)-g(x)\ge \langle \xi ,v\rangle -\sigma \Vert v\Vert ,\quad \text {for all }v\in B(0,\delta ). \end{aligned}$$
Recall that \(f\) satisfies the center Lipschitz condition on \(X\) at \(x\), if there exists \(L>0\) such that
The support function of a set \(K\subset X\) is defined by
The indicator function \(I_S\) of \(S\) is defined by
In view of [14, Proposition 2.2], we have the following result.
Proposition 2.1
\(T(x,s)=0\) if and only if \(x=s\).
We use \(S_0\) to denote the set of all points \(x\in X\) such that \(x\) is a solution of the perturbed optimization problem, i.e.,
Remark 2.1
It is obviously that, if \(f=I_S\), then \(S_0\) equals \(S\) in [14]; if \(f=I_S+J\) and \(U\) is the unit ball in \(X\), then \(S_0\) equals \(S_0\) in [23].
3 Fréchet subdifferential of a minimal time function
Theorem 3.1
Let \(x\in S_0\). The following assertions hold.
-
1.
\(\partial ^F T^f(x)\subset \partial ^F f(x)\cap \{\xi \in X^*{:}\,\mathfrak I _{U}(-\xi )\le 1\}.\)
-
2.
If \(f(\cdot )\) is center Lipschitz on \(X\) at \(x\) with Lipschitz constant \(0\le L<1/M\), where \(M:=\sup _{u\in U}\Vert u\Vert \), then we have
$$\begin{aligned} \partial ^F T^f(x)=\partial ^F f(x)\cap \{\xi \in X^*{:}\,\mathfrak I _{U}(-\xi )\le 1\}. \end{aligned}$$
Proof
(1) Let \(\xi \in \partial ^F T^f(x)\). Then for any \(\sigma >0\), there exists \(\delta >0\) such that
for all \(y\in B(x;\delta )\).
We will prove
Thus \(\xi \in \partial ^Ff(x).\)
By (3.1) and definition of \(S_0\), (3.2) is trivial if \(y\in B(x;\delta )\cap S_0\), we may assume that \(y\in B(x;\delta ){\setminus } S_0\), by the definition of \(T^f\), we have \(T^f(y)\le f(y)\), and as \(x\in S_0\), we have from (3.1) that
Hence, \(f(y)-f(x)-\langle \xi ,y-x\rangle \ge -\sigma \Vert y-x\Vert \), for all \(y\in B(x;\delta )\).
Fix any \(v\in U\). Let \(t_\lambda :=T^f(x-\lambda v)\), where \(\lambda >0\). Since \(x-(x-\lambda v)\in \lambda U\), \(T(x-\lambda v,x)\le \lambda ,\) \(t_\lambda \le \lambda +f(x)<\infty \). It follows from (3.1) that for sufficiently small \(\lambda >0\),
which implies that \(\langle -\xi ,v\rangle \le 1+\sigma \Vert v\Vert \). Since \(\sigma >0\) and \(v\in U\) are arbitrary, \(\mathfrak I _{U}(-\xi )\le 1\).
(2) It is sufficient to prove
Let \(\xi \in \partial ^Ff(x)\) be such that \(\mathfrak I _{U}(-\xi )\le 1\).
For any \(\sigma >0\), take \(\sigma _0\in \left( 0,\frac{(1-LM)\sigma }{(1+M\Vert \xi \Vert )}\right) \). By the definition of Fréchet normal cone, there exists \(\delta >0\) such that
Then
Let \(\delta _0:=\frac{(1-LM)\delta }{3(1+M\Vert \xi \Vert )}< \delta \). Then
Now we prove that (3.5) holds for all \(y\in B(x;\delta _0){\setminus } S_0\). Therefore, \(\xi \in \partial ^FT^f(x)\).
If not, then there is \(y_0\not \in S_0\) such that
The latter implies that
Let \(t:=T^f(y_0)\). By the definition of \(T^f\), for any \({{\mathrm{\varepsilon }}}\in \left( 0,\frac{(1-LM)\delta }{3M}\right) \), there are \(t_1\in (0,t+{{\mathrm{\varepsilon }}})\), and \(s\in X\) such that \(t_1=T(y_0,s)+f(s)<t+{{\mathrm{\varepsilon }}}\). By the definition of \(T\), for any \({{\mathrm{\varepsilon }}}'\in \left( 0,\frac{(1-LM)\delta }{3M}\right) \), there are \(t_2\in (t_1-f(s),t_1-f(s)+{{\mathrm{\varepsilon }}}')\), \(u\in U\), such that \(s-y_0=t_2u.\) Thus (3.7) and \(f\) is center Lipshitz on \(X\) at \(x\) yield that
Then, we have
This verifies that \(s\in B(x;\delta )\). Applying (3.3), (3.8) and \(\mathfrak I _{U}(-\xi )\le 1\), we have
Letting \({{\mathrm{\varepsilon }}}'\rightarrow 0\!+\text { and }{{\mathrm{\varepsilon }}}\rightarrow 0+\), it yields that
which contradicts to (3.6). \(\square \)
In particular, letting \(f=I_S\), we get the following corollary, which was proved in [14].
Corollary 3.1
Assume that \(f=I_S\), where \(S\) is a closed convex subset of \(X\), if \(x\in S\), then
In particular, letting \(f=I_S+J\) and \(U\equiv B\), we get the following corollary, which was proved in [23].
Corollary 3.2
Assume that \(f=I_S+J\) and \(U\equiv B\), where \(B\) is the unit ball in \(X\) and \(S\) is a closed convex subset of \(X\), let \(x\in S_0\). The following assertions hold.
-
1.
\(\partial ^F T^f(x)=\partial ^F d_S^J(x)\subset \partial ^F(J+I_S)(x)\cap B^*.\)
-
2.
If \(J(\cdot )\) is center Lipschitz on \(S\) at \(x\) with Lipschitz constant \(0\le L<1\), then we have
$$\begin{aligned} \partial ^F T^f(x)=\partial ^F d_S^J(x)=\partial ^F(J+I_S)(x)\cap B^*. \end{aligned}$$
4 Proximal subdifferential of a minimal time function
Theorem 4.1
Let \(x\in S_0\). The following assertions hold.
-
1.
\(\partial ^P T^f(x)\subset \partial ^Pf(x)\cap \{\xi \in X^*{:}\,\mathfrak I _{U}(-\xi )\le 1\}.\)
-
2.
If \(f(\cdot )\) is center Lipschitz on \(X\) at \(x\) with Lipschitz constant \(0\le L<1/M\), where \(M:=\sup _{u\in U}\Vert u\Vert \), then we have
$$\begin{aligned} \partial ^P T^f(x)=\partial ^Pf(x)\cap \{\xi \in X^*{:}\,\mathfrak I _{U}(-\xi )\le 1\}. \end{aligned}$$
Proof
(1) Let \(\xi \in \partial ^P T^f(x)\). Then there exist \(\sigma ,\delta >0\) such that
for all \(y\in B(x;\delta )\).
We wil prove
Then \(\xi \in \partial ^Pf(x).\)
By (4.1) and the definition of \(S_0\), (4.2) is trivial if \(y\in B(x;\delta )\cap S_0\), we may assume that \(y\in B(x;\delta )\setminus S_0\). By the definition of \(T^f\), we have \(T^f(y)\le f(y)\), and as \(x\in S_0\), we have from (4.1) that
Hence, \(f(y)-f(x)-\langle \xi ,y-x\rangle \ge -\sigma \Vert y-x\Vert ^2\), for all \(y\in B(x;\delta )\).
Fix any \(v\in U\). Let \(t_\lambda :=T^F(x-\lambda v)\), where \(\lambda >0\). Since \(x-(x-\lambda v)\in \lambda U\), \(T(x-\lambda v,x)\le \lambda ,\) \(t_\lambda \le \lambda +f(x)<\infty \). It follows from (4.1) that for sufficiently small \(\lambda >0\),
which implies that \(\langle -\xi ,v\rangle \le 1\). Therefore, \(\mathfrak I _{U}(-\xi )\le 1\).
(2) It is sufficient to prove
Let \(\xi \in \partial ^Pf(x)\) be such that \(\mathfrak I _{U}(-\xi )\le 1\). Then there exist \(\sigma _1,\delta >0\) such that
Take \(\sigma :=2\left( \frac{1+M\Vert \xi \Vert }{1-LM}\right) ^2\sigma _1>\sigma _1\). Thus (4.3) implies that
and
Let \(\delta _0:=\frac{(1-LM)\delta }{3(1+M\Vert \xi \Vert )}< \delta \). Then
Now we prove that (4.6) holds for all \(y\in B(x;\delta _0){\setminus } S_0\). Therefore, \(\xi \in \partial ^PT^f(x)\).
If not, then there is \(y_0\not \in S_0\) such that
The latter implies that
Let \(t:=T^f(y_0)\). By the definition of \(T^f\), for any \({{\mathrm{\varepsilon }}}\in \left( 0,\frac{(1-LM)\delta }{3M}\right) \), there are \(t_1\in (t,t+{{\mathrm{\varepsilon }}})\), and \(s\in X\) such that \(t_1=T(y_0,s)+f(s)<t+{{\mathrm{\varepsilon }}}\), by the definition of \(T\), for any \({{\mathrm{\varepsilon }}}'\in \left( 0,\frac{(1-LM)\delta }{3M}\right) \), there are \(t_2\in (t_1-f(s),t_1-f(s)+{{\mathrm{\varepsilon }}}')\) \(u\in U\), such that \(s-y_0=t_2u.\) Thus (4.8) and \(f\) is center Lipshitz on \(X\) at \(x\) yield that
Then, we have
This verifies that \(s\in B(x;\delta )\). Applying (4.4), (4.9) and \(\mathfrak I _{U}(-\xi )\le 1\), we have
Letting \({{\mathrm{\varepsilon }}}'\rightarrow 0+\text { and }{{\mathrm{\varepsilon }}}\rightarrow 0+\), it yields that
which contradicts to (4.7). \(\square \)
In particular, letting \(f=I_S\), we get the following corollary, which is proved in [14].
Corollary 4.1
Assume that \(f=I_S\), where \(S\) is a closed convex subset of \(X\), if \(x\in S\), then
In particular, letting \(f=I_S+J\) and \(U\equiv B\), we get the following corollary, which was proved in [23].
Corollary 4.2
Assume that \(f=I_S+J\) and \(U\equiv B\), where \(B\) is the unit ball in \(X\) and \(S\) is a closed convex subset of \(X\), let \(x\in S_0\). The following assertions hold.
-
1.
\(\partial ^P T^f(x)=\partial ^P d_S^J(x)\subset \partial ^P(J+I_S)(x)\cap B^*.\)
-
2.
If \(J(\cdot )\) is center Lipschitz on \(S\) at \(x\) with Lipschitz constant \(0\le L<1\), then we have
$$\begin{aligned} \partial ^P T^f(x)=\partial ^P d_S^J(x)=\partial ^P(J+I_S)(x)\cap B^*. \end{aligned}$$
References
Baranger, J.: Existence de solution pour des problemes doptimisation nonconvexe. C. R. Acad. Sci. Paris 274, 307–309 (1972)
Baranger, J., Temam, R.: Nonconvex optimization problems depending on a parameter. SIAM J. Control 13, 146–152 (1975)
Bidaut, M.F.: Existence theorems for usual and approximate solutions of optimal control problem. J. Optim. Theory Appl. 15, 393–411 (1975)
Burke, J.V., Ferris, M.C., Qian, M.: On the Clarke subdifferential of the distance function of a closed set. J. Math. Anal. Appl. 166(1), 199–213 (1992)
Bounkhel, M., Thibault, L.: On various notions of regularity of sets in nonsmooth analysis. Nonlinear Anal. 48(2, Ser. A: Theory Methods), 223–246 (2002)
Clarke, F.H., Stern, R.J., Wolenski, P.R.: Proximal smoothness and the lower-\(C^{2}\) property. J. Convex Anal. 2(1–2), 117–144 (1995)
Clarke, F.H.: Optimization and Nonsmooth Analysis. Wiley, New York (1983). (a Wiley-Interscience Publication)
Cobzas, S.: Nonconvex optimization problems on weakly compact subsets of Banach spaces. Anal. Numér. Théor. Approx. 9, 19–25 (1980)
Cobzas, S.: Generic existence of solutions for some perturbed optimization problems. J. Math. Anal. Appl. 243, 344–356 (2000)
Colombo, G., Wolenski, P.R.: The subgradient formula for the minimal time function in the case of constant dynamics in Hilbert space. J. Global Optim. 28(3–4), 269–282 (2004)
Colombo, G., Wolenski, P.R.: Variational analysis for a class of minimal time functions in Hilbert spaces. J. Convex Anal. 11(2), 335–361 (2004)
Dontchev, A.L., Zolezzi, T.: Well posed optimization problems. In: Lecure Notes in Mathematics, vol. 1543. Springer, Berlin (1993)
He, Y., Ng, K.F.: Subdifferentials of a minimum time function in Banach spaces. J. Math. Anal. Appl. 321(2), 896–910 (2006)
Jiang, Y., He, Y.: Subdifferentials of a minimum time function in normed spaces. J. Math. Anal. Appl. 358(2), 410–418 (2009)
Lebourg, G.: Perturbed optimization problems in Banach spaces. Bull. Soc. Math. France 60, 95–111 (1979)
Li, C., Peng, L.H.: Porosity of perturbed optimization problems in Banach spaces. J. Math. Anal. Appl. 324, 751–761 (2006)
Meng, L., Li, C., Yao, J.C.: Limiting subdifferentials of perturbed distance functions in Banach spaces. Nonlinear Anal. 75, 1483–1495 (2012)
Ni, R.X.: Generic solutions for some perturbed optimization problem in non-reflexive Banach space. J. Math. Anal. Appl. 302, 417–424 (2005)
Peng, L.H., Li, C., Yao, J.C.: Well-posedness of a class of perturbed optimization problems in Banach spaces. J. Math. Anal. Appl. 346, 384–394 (2008)
Peng, L.H., Li, C.: Existence and porosity for a class of perturbed optimization problems in Banach spaces. J. Math. Anal. Appl. 325, 987–1002 (2007)
Peng, L.H., Li, C., Yao, J.C.: Generic well-posedness for perturbed optimization problems in Banach spaces. Taiwan. J. Math. 14, 1351–1369 (2010)
Stechkin, S.B.: Approximative properties of sets in linear normed spaces. Rev. Math. Pures Appl. 8, 5–18 (1963)
Wang, J.H., Li, C., Xu, H.K.: Subdifferentials of perturbed distance function in Banach spaces. J. Global Optim. 46(4), 489–501 (2010)
Acknowledgments
This work was partially supported by National Natural Science Foundation of China (No. 11271274, No. 11126336 and No. 11201324) and New Teacher’s Fund for Doctor Stations, Ministry of Education (No. 20115134120001).
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Zhang, Y., He, Y. & Jiang, Y. Subdifferentials of a perturbed minimal time function in normed spaces. Optim Lett 8, 1921–1930 (2014). https://doi.org/10.1007/s11590-013-0689-3
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11590-013-0689-3